A randomised subsampling method for change detection

Abstract

In this paper we introduce a method for change detection which uses randomised subsamples of the data. The method is based on the LSCR (Leave-out Sign-dominant Correlation Regions) algorithm for finite sample system identification which generates a region in parameter space which has a guaranteed probability of containing the true parameter. The change detection problem is formulated as a hypothesis testing problem, and the null-hypothesis is accepted if the parameter representing the hypothesis belongs to the confidence set constructed by the LSCR algorithm. This approach delivers a test with a guaranteed low probability of a false alarm for any finite number of observed data points. The te..